Ordinary programs can be proved; the multiple is a same-machine average
The Microsoft Research post is dated October 15, with a page time of 16:00 UTC. Magnetar is a zero-knowledge virtual machine. The opening says it proves programs written in Rust, Go, or C, runs 10 to 39 times faster on average than the prior fastest comparable systems, and rests its security only on standard hash functions.
The measurements later in the post are more specific. On a laptop it proves 7.2 to 8.7 million processor instructions per second. Across 15 benchmarks, from cryptographic hashing to token transfers inside a small Ethereum virtual-machine interpreter, it is 10 to 39 times faster on average than the fastest existing zero-knowledge virtual machines on the same machine. On the token-transfer benchmark it is more than 300 times faster than two others. Those two are not named. An optimized version proves 9.5 to 12.8 million cycles per second, which the post says increases each reported speedup by about 1.3 times. Proofs verify in about 12 milliseconds.
[1]
It proves compiled RISC-V, not a hand-written circuit
A developer writes in Rust, Go, or C, compiles as usual, and Magnetar proves that execution on RISC-V, an open processor with mature compilers. The older approach the post describes is rewriting the program by hand as a cryptographic circuit, then rebuilding the circuit for every change. This system does not ask for that step.
Anyone who receives the short proof can check the result without rerunning the program and without trusting whoever ran it. The zero-knowledge property means the proof does not reveal inputs that should stay private. The post does not turn the millisecond check into a guarantee for every program on every machine. Twelve milliseconds is the verification time in this measurement.
[1]Standard hashes only; an agent proof is still a possibility
Security depends on a standard hash such as SHA-256 or SHA-3. It requires no elliptic curves, no lattices, no special-purpose hash functions, no trusted setup, and no unproven conjecture about error-correcting codes. The post frames that as a choice related to a post-quantum migration: hashes are primitives that have already been studied for a long time. That is the scope of the security assumption, not a test against a quantum computer.
Uses named in the post include an agent attaching a proof that each action complied with the policy it was given, and a regulated financial institution proving that transactions were processed correctly without making the transactions public. Those are written as things that become possible once proofs are routine, not as a shipped product. A technical paper has the design and the measurements. This blog is not the paper.
[1]